Dynamic analysis of malicious behavior propagation based on feature selection in software network

Abstract

In the era of big data, the propagation of malicious software poses a significant threat to corporate data security. To safeguard data assets from the encroachment of malware, it is essential to conduct a dynamic analysis of various information propagation behaviors within software. This paper introduces a dynamic analysis detection method for malicious behavior based on feature extraction (MBDFE), designed to effectively identify and thwart the spread of malicious software. The method is divided into three stages: First, variable-length N-gram algorithms are utilized to extract subsequences of varying lengths from the sample APl call sequences as continuous dynamic features. Second, feature selection techniques based on information gain are employed to identify suitable classification features. Lastly, recurrent neural networks (RNN) are applied for the classification training and prediction of diverse software behaviors. Experimental results and analysis demonstrate that this approach can accurately detect and promptly interrupt the information dissemination of malicious software when such behavior occurs, thereby enhancing the precision and timeliness of malware detection.